Aussie, deep-faked Trudeau a troubling look into future of AI disinfo

Article content

An amusingly awful Facebook ad featuring a heavily accented Justin Trudeau hawking a cryptocurrency scam is a troubling harbinger of things to come as AI blurs the line between real and fake, say technology watchers.

The ad uses apparently AI-generated voices to transform an interview with Prime Minister Justin Trudeau on a 2023 CBC podcast into a three-minute long ad for an online investment scam. Inexplicably, the fake version of the Ottawa-born Trudeau is made to speak with an Australian accent.

Article content

Article content

The ad frames the scam as a government investment program, open only to Canadian citizens, with limited enrollment space. While that may seem nonsensical to savvy users, its low believability is more of a feature rather than a bug, said McGill University assistant professor Aengus Bridgman.

“A trademark of scams is that they need to be realistic enough to catch somebody, but also fake enough so that the people they catch would plausibly go through with (falling for it),” said Bridgman, who studies political communication.

The low quality, he explained, filters out experienced users in favour of the scammers’ target audience: naive and inexperienced individuals willing to spend money on a scheme they deem believable.

“This one is just very poorly done,” Bridgman said of the Trudeau ad, particularly for a scam targeting Canadian citizens.

“But that’s the type of person you want to catch with these ads: somebody who is not digitally literate — in a similar way the elderly in Canada are preyed upon by phone scams and identity theft.”

While the Trudeau video could end up an amusing failure, security business people say government leaders need to be concerned at how fast this technology is progressing.

Article content

Article content

Article content

Tony Anscombe, an executive at cybersecurity firm ESET, said that cinema-grade computer-imaged deepfakes still require trained artists using specialized software. This would be the level of technology used to resurrect deceased celebrities like Peter Cushing in the Star Wars prequel Rogue One. But he said tools capable of creating at least passable fake videos are becoming more available.

This year is set to be a busy election year globally, including high-stakes federal ballots in the United States and India. Bridgman expects heavy use of deepfakes to both interfere and obfuscate in elections and further erode people’s ability to trust what they see online.

“This is going to pollute the information environment in a similar way that misinformation, hyperpolarization and semi-truth already pollute the social media environment,” Bridgman said.

“The consequence of that content is an overall and sustained decrease of trust of all information that the individual has consumed.”

Anscombe said that as technology improves, spotting deepfakes is only going to get more difficult.

Article content

“Because the technology is becoming more available and becoming better, I think that’s becoming more challenging,” he said.

“Then it becomes about where you should be getting your news.”

With truthful online content becoming harder to identify, Anscombe said the best defence is to verify against a trusted source.

Deepfakes have already been cited in one misinformation attempt linked to last month’s U.S. presidential primaries.

In January, Democratic voters in New Hampshire were greeted with robocalls featuring the AI-generated voice of U.S. President Joe Biden, urging people not to vote in the state’s upcoming primaries and to “save their vote” for the November election.

In Russian-speaking Telegram channels this year, deepfaked video depicted the Western-friendly president of Moldova running for re-election this year, Maia Sandu, denigrating herself.

Political deepfakes, Anscombe said, aren’t always used for malicious purposes.

Leading up to last month’s general elections in Pakistan, former president Imran Khan made extensive use of AI-generated video to campaign and give speeches, all while serving a three-year prison sentence.

Article content

Article content

Part of those efforts included a late-December online campaign rally for Khan’s Pakistan Tehreek-e-Insaf (PTI) party, which featured a four-minute AI-generated Khan addressing his supporters.

Bridgman said that technology research is trying to keep up to better identify deepfakes, but said it’s still up to the end user to determine what’s real and what isn’t.

“This is a problem that, to a certain extent defies current resourcing and capabilities of civil society to effectively monitor and respond to,” he said.

“It’s a real challenge right now.”

National Post

bpassifiume@postmedia.com

Our website is the place for the latest breaking news, exclusive scoops, longreads and provocative commentary. Please bookmark nationalpost.com and sign up for our politics newsletter, First Reading, here.

Article content